After The Forecast

Every weather app shows you a forecast. Almost none of them tell you, afterwards, whether it was right. After The Forecast does the checking. Once a month we take what the major forecast models predicted, compare it hour by hour against independent observations, and publish the result for one country at a time.

The scoring runs continuously and covers the world. The reports pick one country a month and look at it properly: which model read the measured temperature and wind best, how fast each one loses its grip as you look further ahead, and exactly how much evidence supports the result. A metric is held back when its observation source is not independent.

How this is measured

We store what each model predicted for a given hour, at distances from one to seven days ahead, and keep it until that hour arrives. Then we compare the prediction against the observation. The first edition uses complete temperature and wind records from airport weather stations, then intersects the sample so every model faces the exact same hours. Rain is held back until independently collected radar observations support a full month.

Every edition states its own limits, in the report, next to the numbers. The methodology explains the scoring in full, and corrections lists anything we have had to fix after publishing.